Design Of Identification System For Blended Oil Products Based On Deep Belief Network

With the rapid development of social economy,refined oil not only occupies an increasingly important position in our daily life,but also plays an important role in the international community,and all countries regard it as a strategic reserve material.While we are highly dependent on the refined oil,we also have higher requirements for its quality.Even if the same but different types of refined oil are blended,the quality of the refined oil will be reduced.Therefore,it is of great significance to identify the blended refined oil quickly.In view of the influence of external noise and other interference factors in the process of collecting spectral data of refined oil products,the methods of multivariate scattering correction,standard normal transformation and derivative of spectral data are used to preprocess the original spectral data of refined oil samples,and the noise reduction effects of these three different preprocessing methods are compared and analyzed.Then,the principal component analysis algorithm and the t-distributed stochastic neighbor embedding algorithm are used to extract the feature information of the preprocessed spectral data,and the clustering effect of the spectral data processed by multivariate scattering correction and t-distributed stochastic neighbor embedding algorithm is more obvious through its feature vector visualization diagram,which provides a reliable data basis for the next model establishment.Finally,the extracted spectral feature data is used as the input data to build the model,and the depth belief network model and the extreme learning machine model are constructed respectively.The research results show that the recognition accuracy of the depth belief network model in the test set is 92.5%,while the recognition accuracy of the extreme learning machine model in the test set is 80%.Therefore,a mixed oil product identification and classification model based on multivariate scattering correction-t-distributed stochastic neighbor embedding algorithm-depth belief network is constructed.According to this model,a complete mixed oil product identification and classification system is designed by using the toolbox of MATLAB GUI,which includes raw spectral data reading module,spectral data preprocessing module,spectral data dimensionality reduction module and identification and classification module.By running the related functions of interface,the goal of rapid and nondestructive identification and classification of mixed oil products is achieved.The research results obtained in this paper can lay a theoretical and technical foundation for the development of rapid identification and classification equipment for mixed oil products.